The six-volume proceedings set LNAI 15919, 15920, 15921, 15922, 15923 and 15924 constitutes the refereed proceedings of the 18th International Conference on Knowledge Science, Engineering and Management, KSEM 2025, held in Macao, China during August 47, 2025.

The 106 papers and 66 short papers are included in these proceedings were carefully reviewed and selected from 354 submissions. They focus on all aspects of the exchange of research in artificial intelligence, data science, knowledge engineering, AI safety, large language models, and related frontier areas.

Klappentext

.- Label Inference Attacks against Federated Unlearning. .- LVLM-FDA: Protecting Large Vision-language models via Fast Detection of Malicious Attempts. .- KAD: Based on the Kolmogorov-Arnold formula’s Nonlinear Dynamic Channel Weights module. .- CoTSentry: Advanced Network Attack Detection with Chain-of-Thought Reasoning. .- A Large Language Model Agent-Guided Multi-Agent System for Adaptive Traffic Signal Control. .- FedBCE: Rethinking Clustered Federated Learning for Better Clustering Efficiency. .- SAML: A Structure-Aware Enhanced Meta Learning Framework for Spatio-Temproal Graph Few-Shot Learning. .- A Task-Specific Feature Fusion Strategy For Small Object Detection. .- Decoupled Dynamic Spatio-Temporal Neural Radiance Fields for Enhanced 3D Scene Synthesis. .- Mamba Model Based on GloVe Word Embedding for Sentiment Analysis. .- DKCER-Agent: Lightweight and Efficient KNN-Enhanced Dynamic Context Optimization for Stepwise Retrieval Augmented Generation. .- Robust AI-Synthesized Image Detection via Multi-feature Frequency-aware Learning. .- Do Domain-Specific LLMs Keep Secrets? An Empirical Study of Privacy Risks and Membership Inference Attacks. .- Dynamic Weighted Consensus Framework for LLM Multi-Agent Debate. .- Streaming Hierarchical Clustering for Emerging New Class. .- Masked Aggregation Learning for Enhancing Distributed Gradient Boosting Decision Trees. .- Personalized learning resource recommendation framework based on knowledge graph and large language model. .- The REGEN model, a LLM-based integrated framework of retrieval and QA generation from plain texts on international freight domain. .- SWV: A Large-scale Sensitive Word Variants Dataset for Semantic Text Matching. .- Explainable Recommendation Using Global Preference Paths on Knowledge Graph. .- Hierarchical Data Protection Based on Homomorphic Encryption Algorithm. .- MCAN-BIFT: An IIoT Intrusion Detection System Integrating Multi-Scale Feature Enhancement and Transformer. .- Summary of the application of deep learning in fault detection. .- Community Detection Attack Based on Balanced Budget Allocation. .- Depth State Space Model for Light Field Depth Estimation via Text-similar Representation. .- GSC-SAGE: A Generative Subgraph Contrastive Framework for Encrypted Traffic Detection. .- Gradient Balanced Part-Whole Relational Weakly Supervised Semantic Segmentation. .- Hierarchical Integration Knowledge Distillation: Enhancing Adversarial Robustness of Student Models via Clean Data Distillation. .- Tree-based Approach for Time-independent Diffusion Network Inference. .- Diffusion model-based multi-scale feature and timing consistency enhancement for ECG signal generation. .- Large Language Models are Not Stable Recommender Systems: A Position Bias Perspective. .- POAgent: A Multi-Agent Controller Towards Adaptive Parameter Optimization.

Inhalt

.- Label Inference Attacks against Federated Unlearning.
.- LVLM-FDA: Protecting Large Vision-language models via Fast Detection of Malicious Attempts.
.- KAD: Based on the Kolmogorov-Arnold formula's Nonlinear Dynamic Channel Weights module.
.- CoTSentry: Advanced Network Attack Detection with Chain-of-Thought Reasoning.
.- A Large Language Model Agent-Guided Multi-Agent System for Adaptive Traffic Signal Control.
.- FedBCE: Rethinking Clustered Federated Learning for Better Clustering Efficiency.
.- SAML: A Structure-Aware Enhanced Meta Learning Framework for Spatio-Temproal Graph Few-Shot Learning.
.- A Task-Specific Feature Fusion Strategy For Small Object Detection.
.- Decoupled Dynamic Spatio-Temporal Neural Radiance Fields for Enhanced 3D Scene Synthesis.
.- Mamba Model Based on GloVe Word Embedding for Sentiment Analysis.
.- DKCER-Agent: Lightweight and Efficient KNN-Enhanced Dynamic Context Optimization for Stepwise Retrieval Augmented Generation.
.- Robust AI-Synthesized Image Detection via Multi-feature Frequency-aware Learning.
.- Do Domain-Specific LLMs Keep Secrets? An Empirical Study of Privacy Risks and Membership Inference Attacks.
.- Dynamic Weighted Consensus Framework for LLM Multi-Agent Debate.
.- Streaming Hierarchical Clustering for Emerging New Class.
.- Masked Aggregation Learning for Enhancing Distributed Gradient Boosting Decision Trees.
.- Personalized learning resource recommendation framework based on knowledge graph and large language model.
.- The REGEN model, a LLM-based integrated framework of retrieval and QA generation from plain texts on international freight domain.
.- SWV: A Large-scale Sensitive Word Variants Dataset for Semantic Text Matching.
.- Explainable Recommendation Using Global Preference Paths on Knowledge Graph.
.- Hierarchical Data Protection Based on Homomorphic Encryption Algorithm.
.- MCAN-BIFT: An IIoT Intrusion Detection System Integrating Multi-Scale Feature Enhancement and Transformer.
.- Summary of the application of deep learning in fault detection.
.- Community Detection Attack Based on Balanced Budget Allocation.
.- Depth State Space Model for Light Field Depth Estimation via Text-similar Representation.
.- GSC-SAGE: A Generative Subgraph Contrastive Framework for Encrypted Traffic Detection.
.- Gradient Balanced Part-Whole Relational Weakly Supervised Semantic Segmentation.
.- Hierarchical Integration Knowledge Distillation: Enhancing Adversarial Robustness of Student Models via Clean Data Distillation.
.- Tree-based Approach for Time-independent Diffusion Network Inference.
.- Diffusion model-based multi-scale feature and timing consistency enhancement for ECG signal generation.
.- Large Language Models are Not Stable Recommender Systems: A Position Bias Perspective.
.- POAgent: A Multi-Agent Controller Towards Adaptive Parameter Optimization.